中国机械工程
中國機械工程
중국궤계공정
China Mechanical Engineering
2015年
20期
2778-2783
,共6页
主成分分析%支持向量机%特征融合%故障诊断%滚动轴承
主成分分析%支持嚮量機%特徵融閤%故障診斷%滾動軸承
주성분분석%지지향량궤%특정융합%고장진단%곤동축승
principal component analysis (PCA)%support vector machine(SVM)%feature fusion%fault diagnosis%rolling bearing
为有效降低滚动轴承故障特征的维数并提高诊断准确率，将主成分分析(PCA)和支持向量机(SVM)方法应用到轴承故障特征的融合分析中，给出了相应的决策流程。应用基于小波包分解的特征提取算法及特征向量的构造方法对不同状态下的振动信号进行分解，得到用于表征轴承运行状态的8维特征集合；应用 PCA 提取累积贡献率达到95%的特征主成分并输入 SVM 分类器中进行识别。结果表明，将滚动轴承故障特征从8维降低到5维，仍可有效表征轴承的状态，但大大降低了计算的复杂性；故障诊断的准确率达到97%以上，诊断时间也相对较短；4种轴承状态识别的准确率从高到低依次为正常、外圈剥落、滚动体剥落和内圈剥落，可为确保设备安全运行和快速故障诊断提供理论依据。
為有效降低滾動軸承故障特徵的維數併提高診斷準確率，將主成分分析(PCA)和支持嚮量機(SVM)方法應用到軸承故障特徵的融閤分析中，給齣瞭相應的決策流程。應用基于小波包分解的特徵提取算法及特徵嚮量的構造方法對不同狀態下的振動信號進行分解，得到用于錶徵軸承運行狀態的8維特徵集閤；應用 PCA 提取纍積貢獻率達到95%的特徵主成分併輸入 SVM 分類器中進行識彆。結果錶明，將滾動軸承故障特徵從8維降低到5維，仍可有效錶徵軸承的狀態，但大大降低瞭計算的複雜性；故障診斷的準確率達到97%以上，診斷時間也相對較短；4種軸承狀態識彆的準確率從高到低依次為正常、外圈剝落、滾動體剝落和內圈剝落，可為確保設備安全運行和快速故障診斷提供理論依據。
위유효강저곤동축승고장특정적유수병제고진단준학솔，장주성분분석(PCA)화지지향량궤(SVM)방법응용도축승고장특정적융합분석중，급출료상응적결책류정。응용기우소파포분해적특정제취산법급특정향량적구조방법대불동상태하적진동신호진행분해，득도용우표정축승운행상태적8유특정집합；응용 PCA 제취루적공헌솔체도95%적특정주성분병수입 SVM 분류기중진행식별。결과표명，장곤동축승고장특정종8유강저도5유，잉가유효표정축승적상태，단대대강저료계산적복잡성；고장진단적준학솔체도97%이상，진단시간야상대교단；4충축승상태식별적준학솔종고도저의차위정상、외권박락、곤동체박락화내권박락，가위학보설비안전운행화쾌속고장진단제공이론의거。
To effectively reduce the dimension of rolling bearing fault features and improve the ac-curacy of diagnosis,the PCA and SVM were applied in the fusion of bearing fault features,and the cor-responding decision-making process was presented.By using the fault feature extraction algorithm and eigenvector constructing methods which were proposed based on wavelet packet decomposition,the bearing vibration signals in different states were decomposed to get the 8-dimensional feature sets which could be used to characterize the running conditions of the bearing.The cumulative contribution rate of 95% principal components were extracted by using PCA method and were input into SVM clas-sifier for identification.Results show that the fault feature dimensions of rolling bearing can be re-duced from 8-dimensions to 5-dimensions,which can still characterize the bearing status effectively, and the computational complexity can be reduced.The fault diagnosis accuracy is higher than 97%,and the diagnosis time is short relatively.The identification accuracy of four bearing status from high to low in turn is normal,outer ring peel,roller peel and inner ring peel.It can ensure the safe operation of the equipment and provide theoretical basis for fast fault diagnosis.
